A recent analysis by Nokia forecasts an impressive increase in wide area network (WAN) traffic, potentially surging by 700% by 2034. This significant growth is primarily attributed to the expansive capabilities of artificial intelligence (AI), especially through inference and agentic systems. Such advancements are revolutionizing data movement across networks. Intriguingly, Nokia estimates that AI could contribute to 30% of all global traffic.
Cameron Daniel, Chief Technology Officer at Megaport, highlighted the transformative nature of AI on digital infrastructure. Unlike previous technological booms such as video streaming, which primarily increased traffic at the network’s edge, AI multiplies demand across all digital layers. The interactions between users and AI systems, plus repetitive cross-regional data transfers, generate unparalleled amounts of data.
Daniel pointed out that although some network routes might experience surges up to tenfold, others may witness only modest changes. Additionally, several AI data corridors have already registered significant traffic jumps. The distinction between training and inference processes is crucial in understanding this growth. While training AI involves large-scale bandwidth use, typically centralized and scheduled, inference relies on real-time, low-latency requirements. This is where AI systems are put to practical use, whether through chat interfaces or autonomous enterprise agents.
The burgeoning interest in agentic AI further underlines its potential implications. These systems, projected to grow at a 26% CAGR through 2034, intensify the demand for efficient inference traffic via the WAN. As AI systems begin direct communications with each other, maximizing efficiency will become increasingly paramount. Daniel believes that metrics like “time-to-first-token” will become vital for performance assessments.
Moreover, a shift towards edge inference is anticipated. This involves moving AI closer to end-users, whether in factories, campuses, or industrial settings. Such evolution demands networks to move beyond traditional best-effort connectivity, requiring consistent low latency and rapid failover mechanisms. As AI systems become integral to operational decision-making, these stringent connectivity demands will only intensify.
However, with this evolution come concerns. Daniel warned of the challenges in managing interconnects and data center links, predicting east-west traffic growth could outpace traditional internet demand. He emphasized the importance of network operators and AI providers in preemptively addressing these challenges.
Security considerations, such as safeguarding against rogue AI agents, are gaining traction. Data leakage prevention tools become invaluable as organizations seek to protect their data against potential threats. Daniel concluded by emphasizing the rise of agent-to-agent communication as a transformative change that could redefine WAN design in the coming decade.
This analysis encourages stakeholders to prepare for the shift, ensuring infrastructure is equipped to handle increasing AI-driven demands. The impending changes offer both exciting opportunities and significant challenges for the telecommunications sector.
